1 /*
2  * Network-device interface management.
3  *
4  * Copyright (c) 2004-2005, Keir Fraser
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License version 2
8  * as published by the Free Software Foundation; or, when distributed
9  * separately from the Linux kernel or incorporated into other
10  * software packages, subject to the following license:
11  *
12  * Permission is hereby granted, free of charge, to any person obtaining a copy
13  * of this source file (the "Software"), to deal in the Software without
14  * restriction, including without limitation the rights to use, copy, modify,
15  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
16  * and to permit persons to whom the Software is furnished to do so, subject to
17  * the following conditions:
18  *
19  * The above copyright notice and this permission notice shall be included in
20  * all copies or substantial portions of the Software.
21  *
22  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
23  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
24  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
25  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
26  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
27  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
28  * IN THE SOFTWARE.
29  */
30 
31 #include "common.h"
32 
33 #include <linux/kthread.h>
34 #include <linux/sched/task.h>
35 #include <linux/ethtool.h>
36 #include <linux/rtnetlink.h>
37 #include <linux/if_vlan.h>
38 #include <linux/vmalloc.h>
39 
40 #include <xen/events.h>
41 #include <asm/xen/hypercall.h>
42 #include <xen/balloon.h>
43 
44 #define XENVIF_QUEUE_LENGTH 32
45 
46 /* Number of bytes allowed on the internal guest Rx queue. */
47 #define XENVIF_RX_QUEUE_BYTES (XEN_NETIF_RX_RING_SIZE/2 * PAGE_SIZE)
48 
49 /* This function is used to set SKBFL_ZEROCOPY_ENABLE as well as
50  * increasing the inflight counter. We need to increase the inflight
51  * counter because core driver calls into xenvif_zerocopy_callback
52  * which calls xenvif_skb_zerocopy_complete.
53  */
54 void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
55 				 struct sk_buff *skb)
56 {
57 	skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_ENABLE;
58 	atomic_inc(&queue->inflight_packets);
59 }
60 
61 void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue)
62 {
63 	atomic_dec(&queue->inflight_packets);
64 
65 	/* Wake the dealloc thread _after_ decrementing inflight_packets so
66 	 * that if kthread_stop() has already been called, the dealloc thread
67 	 * does not wait forever with nothing to wake it.
68 	 */
69 	wake_up(&queue->dealloc_wq);
70 }
71 
72 int xenvif_schedulable(struct xenvif *vif)
73 {
74 	return netif_running(vif->dev) &&
75 		test_bit(VIF_STATUS_CONNECTED, &vif->status) &&
76 		!vif->disabled;
77 }
78 
79 static bool xenvif_handle_tx_interrupt(struct xenvif_queue *queue)
80 {
81 	bool rc;
82 
83 	rc = RING_HAS_UNCONSUMED_REQUESTS(&queue->tx);
84 	if (rc)
85 		napi_schedule(&queue->napi);
86 	return rc;
87 }
88 
89 static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
90 {
91 	struct xenvif_queue *queue = dev_id;
92 	int old;
93 
94 	old = atomic_fetch_or(NETBK_TX_EOI, &queue->eoi_pending);
95 	WARN(old & NETBK_TX_EOI, "Interrupt while EOI pending\n");
96 
97 	if (!xenvif_handle_tx_interrupt(queue)) {
98 		atomic_andnot(NETBK_TX_EOI, &queue->eoi_pending);
99 		xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
100 	}
101 
102 	return IRQ_HANDLED;
103 }
104 
105 static int xenvif_poll(struct napi_struct *napi, int budget)
106 {
107 	struct xenvif_queue *queue =
108 		container_of(napi, struct xenvif_queue, napi);
109 	int work_done;
110 
111 	/* This vif is rogue, we pretend we've there is nothing to do
112 	 * for this vif to deschedule it from NAPI. But this interface
113 	 * will be turned off in thread context later.
114 	 */
115 	if (unlikely(queue->vif->disabled)) {
116 		napi_complete(napi);
117 		return 0;
118 	}
119 
120 	work_done = xenvif_tx_action(queue, budget);
121 
122 	if (work_done < budget) {
123 		napi_complete_done(napi, work_done);
124 		/* If the queue is rate-limited, it shall be
125 		 * rescheduled in the timer callback.
126 		 */
127 		if (likely(!queue->rate_limited))
128 			xenvif_napi_schedule_or_enable_events(queue);
129 	}
130 
131 	return work_done;
132 }
133 
134 static bool xenvif_handle_rx_interrupt(struct xenvif_queue *queue)
135 {
136 	bool rc;
137 
138 	rc = xenvif_have_rx_work(queue, false);
139 	if (rc)
140 		xenvif_kick_thread(queue);
141 	return rc;
142 }
143 
144 static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
145 {
146 	struct xenvif_queue *queue = dev_id;
147 	int old;
148 
149 	old = atomic_fetch_or(NETBK_RX_EOI, &queue->eoi_pending);
150 	WARN(old & NETBK_RX_EOI, "Interrupt while EOI pending\n");
151 
152 	if (!xenvif_handle_rx_interrupt(queue)) {
153 		atomic_andnot(NETBK_RX_EOI, &queue->eoi_pending);
154 		xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
155 	}
156 
157 	return IRQ_HANDLED;
158 }
159 
160 irqreturn_t xenvif_interrupt(int irq, void *dev_id)
161 {
162 	struct xenvif_queue *queue = dev_id;
163 	int old;
164 	bool has_rx, has_tx;
165 
166 	old = atomic_fetch_or(NETBK_COMMON_EOI, &queue->eoi_pending);
167 	WARN(old, "Interrupt while EOI pending\n");
168 
169 	has_tx = xenvif_handle_tx_interrupt(queue);
170 	has_rx = xenvif_handle_rx_interrupt(queue);
171 
172 	if (!has_rx && !has_tx) {
173 		atomic_andnot(NETBK_COMMON_EOI, &queue->eoi_pending);
174 		xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
175 	}
176 
177 	return IRQ_HANDLED;
178 }
179 
180 int xenvif_queue_stopped(struct xenvif_queue *queue)
181 {
182 	struct net_device *dev = queue->vif->dev;
183 	unsigned int id = queue->id;
184 	return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
185 }
186 
187 void xenvif_wake_queue(struct xenvif_queue *queue)
188 {
189 	struct net_device *dev = queue->vif->dev;
190 	unsigned int id = queue->id;
191 	netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
192 }
193 
194 static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb,
195 			       struct net_device *sb_dev)
196 {
197 	struct xenvif *vif = netdev_priv(dev);
198 	unsigned int size = vif->hash.size;
199 	unsigned int num_queues;
200 
201 	/* If queues are not set up internally - always return 0
202 	 * as the packet going to be dropped anyway */
203 	num_queues = READ_ONCE(vif->num_queues);
204 	if (num_queues < 1)
205 		return 0;
206 
207 	if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
208 		return netdev_pick_tx(dev, skb, NULL) %
209 		       dev->real_num_tx_queues;
210 
211 	xenvif_set_skb_hash(vif, skb);
212 
213 	if (size == 0)
214 		return skb_get_hash_raw(skb) % dev->real_num_tx_queues;
215 
216 	return vif->hash.mapping[vif->hash.mapping_sel]
217 				[skb_get_hash_raw(skb) % size];
218 }
219 
220 static netdev_tx_t
221 xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
222 {
223 	struct xenvif *vif = netdev_priv(dev);
224 	struct xenvif_queue *queue = NULL;
225 	unsigned int num_queues;
226 	u16 index;
227 	struct xenvif_rx_cb *cb;
228 
229 	BUG_ON(skb->dev != dev);
230 
231 	/* Drop the packet if queues are not set up.
232 	 * This handler should be called inside an RCU read section
233 	 * so we don't need to enter it here explicitly.
234 	 */
235 	num_queues = READ_ONCE(vif->num_queues);
236 	if (num_queues < 1)
237 		goto drop;
238 
239 	/* Obtain the queue to be used to transmit this packet */
240 	index = skb_get_queue_mapping(skb);
241 	if (index >= num_queues) {
242 		pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n",
243 				    index, vif->dev->name);
244 		index %= num_queues;
245 	}
246 	queue = &vif->queues[index];
247 
248 	/* Drop the packet if queue is not ready */
249 	if (queue->task == NULL ||
250 	    queue->dealloc_task == NULL ||
251 	    !xenvif_schedulable(vif))
252 		goto drop;
253 
254 	if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
255 		struct ethhdr *eth = (struct ethhdr *)skb->data;
256 
257 		if (!xenvif_mcast_match(vif, eth->h_dest))
258 			goto drop;
259 	}
260 
261 	cb = XENVIF_RX_CB(skb);
262 	cb->expires = jiffies + vif->drain_timeout;
263 
264 	/* If there is no hash algorithm configured then make sure there
265 	 * is no hash information in the socket buffer otherwise it
266 	 * would be incorrectly forwarded to the frontend.
267 	 */
268 	if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
269 		skb_clear_hash(skb);
270 
271 	xenvif_rx_queue_tail(queue, skb);
272 	xenvif_kick_thread(queue);
273 
274 	return NETDEV_TX_OK;
275 
276  drop:
277 	vif->dev->stats.tx_dropped++;
278 	dev_kfree_skb(skb);
279 	return NETDEV_TX_OK;
280 }
281 
282 static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
283 {
284 	struct xenvif *vif = netdev_priv(dev);
285 	struct xenvif_queue *queue = NULL;
286 	unsigned int num_queues;
287 	u64 rx_bytes = 0;
288 	u64 rx_packets = 0;
289 	u64 tx_bytes = 0;
290 	u64 tx_packets = 0;
291 	unsigned int index;
292 
293 	rcu_read_lock();
294 	num_queues = READ_ONCE(vif->num_queues);
295 
296 	/* Aggregate tx and rx stats from each queue */
297 	for (index = 0; index < num_queues; ++index) {
298 		queue = &vif->queues[index];
299 		rx_bytes += queue->stats.rx_bytes;
300 		rx_packets += queue->stats.rx_packets;
301 		tx_bytes += queue->stats.tx_bytes;
302 		tx_packets += queue->stats.tx_packets;
303 	}
304 
305 	rcu_read_unlock();
306 
307 	vif->dev->stats.rx_bytes = rx_bytes;
308 	vif->dev->stats.rx_packets = rx_packets;
309 	vif->dev->stats.tx_bytes = tx_bytes;
310 	vif->dev->stats.tx_packets = tx_packets;
311 
312 	return &vif->dev->stats;
313 }
314 
315 static void xenvif_up(struct xenvif *vif)
316 {
317 	struct xenvif_queue *queue = NULL;
318 	unsigned int num_queues = vif->num_queues;
319 	unsigned int queue_index;
320 
321 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
322 		queue = &vif->queues[queue_index];
323 		napi_enable(&queue->napi);
324 		enable_irq(queue->tx_irq);
325 		if (queue->tx_irq != queue->rx_irq)
326 			enable_irq(queue->rx_irq);
327 		xenvif_napi_schedule_or_enable_events(queue);
328 	}
329 }
330 
331 static void xenvif_down(struct xenvif *vif)
332 {
333 	struct xenvif_queue *queue = NULL;
334 	unsigned int num_queues = vif->num_queues;
335 	unsigned int queue_index;
336 
337 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
338 		queue = &vif->queues[queue_index];
339 		disable_irq(queue->tx_irq);
340 		if (queue->tx_irq != queue->rx_irq)
341 			disable_irq(queue->rx_irq);
342 		napi_disable(&queue->napi);
343 		del_timer_sync(&queue->credit_timeout);
344 	}
345 }
346 
347 static int xenvif_open(struct net_device *dev)
348 {
349 	struct xenvif *vif = netdev_priv(dev);
350 	if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
351 		xenvif_up(vif);
352 	netif_tx_start_all_queues(dev);
353 	return 0;
354 }
355 
356 static int xenvif_close(struct net_device *dev)
357 {
358 	struct xenvif *vif = netdev_priv(dev);
359 	if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
360 		xenvif_down(vif);
361 	netif_tx_stop_all_queues(dev);
362 	return 0;
363 }
364 
365 static int xenvif_change_mtu(struct net_device *dev, int mtu)
366 {
367 	struct xenvif *vif = netdev_priv(dev);
368 	int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN;
369 
370 	if (mtu > max)
371 		return -EINVAL;
372 	dev->mtu = mtu;
373 	return 0;
374 }
375 
376 static netdev_features_t xenvif_fix_features(struct net_device *dev,
377 	netdev_features_t features)
378 {
379 	struct xenvif *vif = netdev_priv(dev);
380 
381 	if (!vif->can_sg)
382 		features &= ~NETIF_F_SG;
383 	if (~(vif->gso_mask) & GSO_BIT(TCPV4))
384 		features &= ~NETIF_F_TSO;
385 	if (~(vif->gso_mask) & GSO_BIT(TCPV6))
386 		features &= ~NETIF_F_TSO6;
387 	if (!vif->ip_csum)
388 		features &= ~NETIF_F_IP_CSUM;
389 	if (!vif->ipv6_csum)
390 		features &= ~NETIF_F_IPV6_CSUM;
391 
392 	return features;
393 }
394 
395 static const struct xenvif_stat {
396 	char name[ETH_GSTRING_LEN];
397 	u16 offset;
398 } xenvif_stats[] = {
399 	{
400 		"rx_gso_checksum_fixup",
401 		offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
402 	},
403 	/* If (sent != success + fail), there are probably packets never
404 	 * freed up properly!
405 	 */
406 	{
407 		"tx_zerocopy_sent",
408 		offsetof(struct xenvif_stats, tx_zerocopy_sent),
409 	},
410 	{
411 		"tx_zerocopy_success",
412 		offsetof(struct xenvif_stats, tx_zerocopy_success),
413 	},
414 	{
415 		"tx_zerocopy_fail",
416 		offsetof(struct xenvif_stats, tx_zerocopy_fail)
417 	},
418 	/* Number of packets exceeding MAX_SKB_FRAG slots. You should use
419 	 * a guest with the same MAX_SKB_FRAG
420 	 */
421 	{
422 		"tx_frag_overflow",
423 		offsetof(struct xenvif_stats, tx_frag_overflow)
424 	},
425 };
426 
427 static int xenvif_get_sset_count(struct net_device *dev, int string_set)
428 {
429 	switch (string_set) {
430 	case ETH_SS_STATS:
431 		return ARRAY_SIZE(xenvif_stats);
432 	default:
433 		return -EINVAL;
434 	}
435 }
436 
437 static void xenvif_get_ethtool_stats(struct net_device *dev,
438 				     struct ethtool_stats *stats, u64 * data)
439 {
440 	struct xenvif *vif = netdev_priv(dev);
441 	unsigned int num_queues;
442 	int i;
443 	unsigned int queue_index;
444 
445 	rcu_read_lock();
446 	num_queues = READ_ONCE(vif->num_queues);
447 
448 	for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
449 		unsigned long accum = 0;
450 		for (queue_index = 0; queue_index < num_queues; ++queue_index) {
451 			void *vif_stats = &vif->queues[queue_index].stats;
452 			accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
453 		}
454 		data[i] = accum;
455 	}
456 
457 	rcu_read_unlock();
458 }
459 
460 static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
461 {
462 	int i;
463 
464 	switch (stringset) {
465 	case ETH_SS_STATS:
466 		for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
467 			memcpy(data + i * ETH_GSTRING_LEN,
468 			       xenvif_stats[i].name, ETH_GSTRING_LEN);
469 		break;
470 	}
471 }
472 
473 static const struct ethtool_ops xenvif_ethtool_ops = {
474 	.get_link	= ethtool_op_get_link,
475 
476 	.get_sset_count = xenvif_get_sset_count,
477 	.get_ethtool_stats = xenvif_get_ethtool_stats,
478 	.get_strings = xenvif_get_strings,
479 };
480 
481 static const struct net_device_ops xenvif_netdev_ops = {
482 	.ndo_select_queue = xenvif_select_queue,
483 	.ndo_start_xmit	= xenvif_start_xmit,
484 	.ndo_get_stats	= xenvif_get_stats,
485 	.ndo_open	= xenvif_open,
486 	.ndo_stop	= xenvif_close,
487 	.ndo_change_mtu	= xenvif_change_mtu,
488 	.ndo_fix_features = xenvif_fix_features,
489 	.ndo_set_mac_address = eth_mac_addr,
490 	.ndo_validate_addr   = eth_validate_addr,
491 };
492 
493 struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
494 			    unsigned int handle)
495 {
496 	static const u8 dummy_addr[ETH_ALEN] = {
497 		0xfe, 0xff, 0xff, 0xff, 0xff, 0xff,
498 	};
499 	int err;
500 	struct net_device *dev;
501 	struct xenvif *vif;
502 	char name[IFNAMSIZ] = {};
503 
504 	snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
505 	/* Allocate a netdev with the max. supported number of queues.
506 	 * When the guest selects the desired number, it will be updated
507 	 * via netif_set_real_num_*_queues().
508 	 */
509 	dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
510 			      ether_setup, xenvif_max_queues);
511 	if (dev == NULL) {
512 		pr_warn("Could not allocate netdev for %s\n", name);
513 		return ERR_PTR(-ENOMEM);
514 	}
515 
516 	SET_NETDEV_DEV(dev, parent);
517 
518 	vif = netdev_priv(dev);
519 
520 	vif->domid  = domid;
521 	vif->handle = handle;
522 	vif->can_sg = 1;
523 	vif->ip_csum = 1;
524 	vif->dev = dev;
525 	vif->disabled = false;
526 	vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
527 	vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
528 
529 	/* Start out with no queues. */
530 	vif->queues = NULL;
531 	vif->num_queues = 0;
532 
533 	vif->xdp_headroom = 0;
534 
535 	spin_lock_init(&vif->lock);
536 	INIT_LIST_HEAD(&vif->fe_mcast_addr);
537 
538 	dev->netdev_ops	= &xenvif_netdev_ops;
539 	dev->hw_features = NETIF_F_SG |
540 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
541 		NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST;
542 	dev->features = dev->hw_features | NETIF_F_RXCSUM;
543 	dev->ethtool_ops = &xenvif_ethtool_ops;
544 
545 	dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
546 
547 	dev->min_mtu = ETH_MIN_MTU;
548 	dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN;
549 
550 	/*
551 	 * Initialise a dummy MAC address. We choose the numerically
552 	 * largest non-broadcast address to prevent the address getting
553 	 * stolen by an Ethernet bridge for STP purposes.
554 	 * (FE:FF:FF:FF:FF:FF)
555 	 */
556 	eth_hw_addr_set(dev, dummy_addr);
557 
558 	netif_carrier_off(dev);
559 
560 	err = register_netdev(dev);
561 	if (err) {
562 		netdev_warn(dev, "Could not register device: err=%d\n", err);
563 		free_netdev(dev);
564 		return ERR_PTR(err);
565 	}
566 
567 	netdev_dbg(dev, "Successfully created xenvif\n");
568 
569 	__module_get(THIS_MODULE);
570 
571 	return vif;
572 }
573 
574 int xenvif_init_queue(struct xenvif_queue *queue)
575 {
576 	int err, i;
577 
578 	queue->credit_bytes = queue->remaining_credit = ~0UL;
579 	queue->credit_usec  = 0UL;
580 	timer_setup(&queue->credit_timeout, xenvif_tx_credit_callback, 0);
581 	queue->credit_window_start = get_jiffies_64();
582 
583 	queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
584 
585 	skb_queue_head_init(&queue->rx_queue);
586 	skb_queue_head_init(&queue->tx_queue);
587 
588 	queue->pending_cons = 0;
589 	queue->pending_prod = MAX_PENDING_REQS;
590 	for (i = 0; i < MAX_PENDING_REQS; ++i)
591 		queue->pending_ring[i] = i;
592 
593 	spin_lock_init(&queue->callback_lock);
594 	spin_lock_init(&queue->response_lock);
595 
596 	/* If ballooning is disabled, this will consume real memory, so you
597 	 * better enable it. The long term solution would be to use just a
598 	 * bunch of valid page descriptors, without dependency on ballooning
599 	 */
600 	err = gnttab_alloc_pages(MAX_PENDING_REQS,
601 				 queue->mmap_pages);
602 	if (err) {
603 		netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
604 		return -ENOMEM;
605 	}
606 
607 	for (i = 0; i < MAX_PENDING_REQS; i++) {
608 		queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
609 			{ .callback = xenvif_zerocopy_callback,
610 			  { { .ctx = NULL,
611 			      .desc = i } } };
612 		queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
613 	}
614 
615 	return 0;
616 }
617 
618 void xenvif_carrier_on(struct xenvif *vif)
619 {
620 	rtnl_lock();
621 	if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
622 		dev_set_mtu(vif->dev, ETH_DATA_LEN);
623 	netdev_update_features(vif->dev);
624 	set_bit(VIF_STATUS_CONNECTED, &vif->status);
625 	if (netif_running(vif->dev))
626 		xenvif_up(vif);
627 	rtnl_unlock();
628 }
629 
630 int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
631 			unsigned int evtchn)
632 {
633 	struct net_device *dev = vif->dev;
634 	struct xenbus_device *xendev = xenvif_to_xenbus_device(vif);
635 	void *addr;
636 	struct xen_netif_ctrl_sring *shared;
637 	RING_IDX rsp_prod, req_prod;
638 	int err;
639 
640 	err = xenbus_map_ring_valloc(xendev, &ring_ref, 1, &addr);
641 	if (err)
642 		goto err;
643 
644 	shared = (struct xen_netif_ctrl_sring *)addr;
645 	rsp_prod = READ_ONCE(shared->rsp_prod);
646 	req_prod = READ_ONCE(shared->req_prod);
647 
648 	BACK_RING_ATTACH(&vif->ctrl, shared, rsp_prod, XEN_PAGE_SIZE);
649 
650 	err = -EIO;
651 	if (req_prod - rsp_prod > RING_SIZE(&vif->ctrl))
652 		goto err_unmap;
653 
654 	err = bind_interdomain_evtchn_to_irq_lateeoi(xendev, evtchn);
655 	if (err < 0)
656 		goto err_unmap;
657 
658 	vif->ctrl_irq = err;
659 
660 	xenvif_init_hash(vif);
661 
662 	err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn,
663 				   IRQF_ONESHOT, "xen-netback-ctrl", vif);
664 	if (err) {
665 		pr_warn("Could not setup irq handler for %s\n", dev->name);
666 		goto err_deinit;
667 	}
668 
669 	return 0;
670 
671 err_deinit:
672 	xenvif_deinit_hash(vif);
673 	unbind_from_irqhandler(vif->ctrl_irq, vif);
674 	vif->ctrl_irq = 0;
675 
676 err_unmap:
677 	xenbus_unmap_ring_vfree(xendev, vif->ctrl.sring);
678 	vif->ctrl.sring = NULL;
679 
680 err:
681 	return err;
682 }
683 
684 static void xenvif_disconnect_queue(struct xenvif_queue *queue)
685 {
686 	if (queue->task) {
687 		kthread_stop(queue->task);
688 		put_task_struct(queue->task);
689 		queue->task = NULL;
690 	}
691 
692 	if (queue->dealloc_task) {
693 		kthread_stop(queue->dealloc_task);
694 		queue->dealloc_task = NULL;
695 	}
696 
697 	if (queue->napi.poll) {
698 		netif_napi_del(&queue->napi);
699 		queue->napi.poll = NULL;
700 	}
701 
702 	if (queue->tx_irq) {
703 		unbind_from_irqhandler(queue->tx_irq, queue);
704 		if (queue->tx_irq == queue->rx_irq)
705 			queue->rx_irq = 0;
706 		queue->tx_irq = 0;
707 	}
708 
709 	if (queue->rx_irq) {
710 		unbind_from_irqhandler(queue->rx_irq, queue);
711 		queue->rx_irq = 0;
712 	}
713 
714 	xenvif_unmap_frontend_data_rings(queue);
715 }
716 
717 int xenvif_connect_data(struct xenvif_queue *queue,
718 			unsigned long tx_ring_ref,
719 			unsigned long rx_ring_ref,
720 			unsigned int tx_evtchn,
721 			unsigned int rx_evtchn)
722 {
723 	struct xenbus_device *dev = xenvif_to_xenbus_device(queue->vif);
724 	struct task_struct *task;
725 	int err;
726 
727 	BUG_ON(queue->tx_irq);
728 	BUG_ON(queue->task);
729 	BUG_ON(queue->dealloc_task);
730 
731 	err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
732 					     rx_ring_ref);
733 	if (err < 0)
734 		goto err;
735 
736 	init_waitqueue_head(&queue->wq);
737 	init_waitqueue_head(&queue->dealloc_wq);
738 	atomic_set(&queue->inflight_packets, 0);
739 
740 	netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
741 			NAPI_POLL_WEIGHT);
742 
743 	queue->stalled = true;
744 
745 	task = kthread_run(xenvif_kthread_guest_rx, queue,
746 			   "%s-guest-rx", queue->name);
747 	if (IS_ERR(task))
748 		goto kthread_err;
749 	queue->task = task;
750 	/*
751 	 * Take a reference to the task in order to prevent it from being freed
752 	 * if the thread function returns before kthread_stop is called.
753 	 */
754 	get_task_struct(task);
755 
756 	task = kthread_run(xenvif_dealloc_kthread, queue,
757 			   "%s-dealloc", queue->name);
758 	if (IS_ERR(task))
759 		goto kthread_err;
760 	queue->dealloc_task = task;
761 
762 	if (tx_evtchn == rx_evtchn) {
763 		/* feature-split-event-channels == 0 */
764 		err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
765 			dev, tx_evtchn, xenvif_interrupt, 0,
766 			queue->name, queue);
767 		if (err < 0)
768 			goto err;
769 		queue->tx_irq = queue->rx_irq = err;
770 		disable_irq(queue->tx_irq);
771 	} else {
772 		/* feature-split-event-channels == 1 */
773 		snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
774 			 "%s-tx", queue->name);
775 		err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
776 			dev, tx_evtchn, xenvif_tx_interrupt, 0,
777 			queue->tx_irq_name, queue);
778 		if (err < 0)
779 			goto err;
780 		queue->tx_irq = err;
781 		disable_irq(queue->tx_irq);
782 
783 		snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
784 			 "%s-rx", queue->name);
785 		err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
786 			dev, rx_evtchn, xenvif_rx_interrupt, 0,
787 			queue->rx_irq_name, queue);
788 		if (err < 0)
789 			goto err;
790 		queue->rx_irq = err;
791 		disable_irq(queue->rx_irq);
792 	}
793 
794 	return 0;
795 
796 kthread_err:
797 	pr_warn("Could not allocate kthread for %s\n", queue->name);
798 	err = PTR_ERR(task);
799 err:
800 	xenvif_disconnect_queue(queue);
801 	return err;
802 }
803 
804 void xenvif_carrier_off(struct xenvif *vif)
805 {
806 	struct net_device *dev = vif->dev;
807 
808 	rtnl_lock();
809 	if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
810 		netif_carrier_off(dev); /* discard queued packets */
811 		if (netif_running(dev))
812 			xenvif_down(vif);
813 	}
814 	rtnl_unlock();
815 }
816 
817 void xenvif_disconnect_data(struct xenvif *vif)
818 {
819 	struct xenvif_queue *queue = NULL;
820 	unsigned int num_queues = vif->num_queues;
821 	unsigned int queue_index;
822 
823 	xenvif_carrier_off(vif);
824 
825 	for (queue_index = 0; queue_index < num_queues; ++queue_index) {
826 		queue = &vif->queues[queue_index];
827 
828 		xenvif_disconnect_queue(queue);
829 	}
830 
831 	xenvif_mcast_addr_list_free(vif);
832 }
833 
834 void xenvif_disconnect_ctrl(struct xenvif *vif)
835 {
836 	if (vif->ctrl_irq) {
837 		xenvif_deinit_hash(vif);
838 		unbind_from_irqhandler(vif->ctrl_irq, vif);
839 		vif->ctrl_irq = 0;
840 	}
841 
842 	if (vif->ctrl.sring) {
843 		xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
844 					vif->ctrl.sring);
845 		vif->ctrl.sring = NULL;
846 	}
847 }
848 
849 /* Reverse the relevant parts of xenvif_init_queue().
850  * Used for queue teardown from xenvif_free(), and on the
851  * error handling paths in xenbus.c:connect().
852  */
853 void xenvif_deinit_queue(struct xenvif_queue *queue)
854 {
855 	gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
856 }
857 
858 void xenvif_free(struct xenvif *vif)
859 {
860 	struct xenvif_queue *queues = vif->queues;
861 	unsigned int num_queues = vif->num_queues;
862 	unsigned int queue_index;
863 
864 	unregister_netdev(vif->dev);
865 	free_netdev(vif->dev);
866 
867 	for (queue_index = 0; queue_index < num_queues; ++queue_index)
868 		xenvif_deinit_queue(&queues[queue_index]);
869 	vfree(queues);
870 
871 	module_put(THIS_MODULE);
872 }
873